Talk by Manuel Lopes from Inria Bordeaux, France. Given to the Redwood Center for Theoretical Neuroscience at UC Berkeley.
Abstract Curiosity is very predominant in people and many animals, but its mechanisms are poorly understood. Motivated by research on neuroscience and child development we want to develop methods to understand those mechanism and provide machines with adaptation capabilities similar to the ones of animals. Our model includes exploration biases such as reward, information, learning progress and reward anticipation. We show that only the combination those biases can explain the observed empirical data. We also show how exploration biases based on learning progress can improve robustness to lack of knowledge about a domain and its robustness to environmental changes.
